pith. sign in

arxiv: 2605.19595 · v1 · pith:GHWGML3Dnew · submitted 2026-05-19 · 💻 cs.CV · cs.AI

A novel YOLO26-MoE optimized by an LLM agent for insulator fault detection considering UAV images

Jo\~ao Pedro Matos-Carvalho , Laio Oriel Seman , Stefano Frizzo Stefenon , Mohammad Khalaf Mohammad Khreasat , Gabriel Villarrubia Gonz\'alez This is my paper

Pith reviewed 2026-05-20 05:38 UTC · model grok-4.3

classification 💻 cs.CV cs.AI
keywords insulator fault detectionUAV imagesYOLO26Mixture of Expertspower line inspectionobject detectionLLM agent optimizationaerial imagery
0
0 comments X

The pith

Integrating a sparse Mixture-of-Experts module into YOLO26 improves detection of subtle insulator faults in UAV images to 0.99 mAP at 0.5 IoU.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper introduces YOLO26-MoE, which modifies the YOLO26 detector by inserting a sparse Mixture-of-Experts layer into its high-resolution feature path. This change lets the model dynamically select expert networks to better process the small, varied faults on insulators seen in UAV footage. An LLM-based agent handles the hyperparameter search and training process. The result is a detector that reaches 0.99 mean average precision at 0.5 overlap while staying efficient enough for practical use in power grid monitoring.

Core claim

The proposed YOLO26-MoE architecture integrates a sparse Mixture-of-Experts module into the high-resolution branch of the YOLO26 detector to enable adaptive feature refinement for subtle and diverse fault patterns in UAV images of power line insulators, while maintaining the efficiency of a one-stage framework. Optimized via a tool-augmented LLM agent, the model achieves 0.9900 mAP@0.5 and 0.9515 mAP@0.5:0.95, outperforming latest YOLO versions.

What carries the argument

Sparse Mixture-of-Experts (MoE) module integrated into the high-resolution branch of YOLO26, which adaptively refines features for diverse fault patterns without sacrificing one-stage detection speed.

Load-bearing premise

The sparse MoE module can be integrated into the high-resolution branch without disrupting the overall detector's balance between accuracy and computational efficiency.

What would settle it

Running the base YOLO26 without the MoE addition on the same insulator UAV dataset and observing whether the mAP scores fall significantly below 0.99 at 0.5 IoU.

Figures

Figures reproduced from arXiv: 2605.19595 by Gabriel Villarrubia Gonz\'alez, Jo\~ao Pedro Matos-Carvalho, Laio Oriel Seman, Mohammad Khalaf Mohammad Khreasat, Stefano Frizzo Stefenon.

Figure 1
Figure 1. Figure 1: Workflow of the YOLO26-MoE optimization and evaluation process performed by an LLM agent. [PITH_FULL_IMAGE:figures/full_fig_p010_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Comparison between the standard YOLO26 architecture and the proposed YOLO26-MoE variant. [PITH_FULL_IMAGE:figures/full_fig_p012_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Training pipeline of the proposed YOLO26-MoE detector. The workflow combines standard [PITH_FULL_IMAGE:figures/full_fig_p015_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Samples from the original dataset highlighted: a) flashover and b) broken insulators. [PITH_FULL_IMAGE:figures/full_fig_p020_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Optimization history of the Optuna study over 50 trials. Blue markers denote the objective value [PITH_FULL_IMAGE:figures/full_fig_p022_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Relative hyperparameter importance with respect to the validation objective in the Optuna-based [PITH_FULL_IMAGE:figures/full_fig_p023_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Evolution of mAP@0.50 and mAP@0.50:0.95 over 500 training epochs for the final optimized [PITH_FULL_IMAGE:figures/full_fig_p024_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Violin plots of mAP@0.5, mAP@0.5:0.95, precision, recall, and F1-score over 50 runs of the [PITH_FULL_IMAGE:figures/full_fig_p027_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Boxplot comparison of mAP@0.5:0.95 on the test split between the proposed YOLO26-MoE model [PITH_FULL_IMAGE:figures/full_fig_p030_9.png] view at source ↗
read the original abstract

The inspection of electrical power line insulators is essential for ensuring grid reliability and preventing failures caused by damaged or degraded insulation components. In recent years, Unmanned Aerial Vehicles (UAVs) combined with deep learning-based vision systems have emerged as an effective solution for automating this process. However, insulator fault detection remains challenging due to small defect regions, heterogeneous fault patterns, complex backgrounds, and varying imaging conditions. To address these challenges, this paper proposes an optimized YOLO26-MoE, a novel object detection architecture that integrates a sparse Mixture-of-Experts (MoE) module into the high-resolution branch of the YOLO26 detector. The proposed modification enables adaptive feature refinement for subtle and diverse fault patterns while preserving the efficiency of a one-stage detection framework. Hyperparameter optimization, final training, and evaluation were coordinated through a tool-augmented Large Language Model (LLM) agent. The proposed model achieved 0.9900 mAP@0.5 and 0.9515 mAP@0.5:0.95, outperforming the latest YOLO versions. These results demonstrate that the proposed model provides an effective and reliable solution for UAV-based insulator fault detection.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 1 minor

Summary. The paper proposes YOLO26-MoE, a modification of the YOLO26 one-stage detector that inserts a sparse Mixture-of-Experts module into the high-resolution branch to enable adaptive feature refinement for small and diverse insulator faults in UAV imagery. Hyperparameter search, training, and evaluation are performed via a tool-augmented LLM agent. The manuscript reports final performance of 0.9900 mAP@0.5 and 0.9515 mAP@0.5:0.95 and states that these scores exceed those of the latest YOLO variants.

Significance. If the reported gains can be reproduced and attributed to the MoE insertion plus LLM-driven optimization, the work would supply a practical, efficient detector for a high-value industrial inspection task. The combination of sparse MoE with a high-resolution branch and LLM-based tuning is a plausible route to handling heterogeneous fault patterns without sacrificing one-stage speed; however, the current evidence does not yet isolate these contributions.

major comments (3)
  1. Abstract: the central performance claim (0.9900 mAP@0.5, 0.9515 mAP@0.5:0.95, outperforming latest YOLO versions) is presented without any dataset description, train/test split, baseline implementations, ablation tables, or error bars. Because the attribution of these scores to the MoE module and LLM agent is the load-bearing assertion, the absence of controlled comparisons prevents verification of the claimed improvements.
  2. Method section (description of MoE integration): the sparse MoE module is stated to be placed in the high-resolution branch, yet no equations, routing function, expert count, or capacity factor are supplied. Without these details it is impossible to assess whether the modification is reproducible or whether it genuinely preserves one-stage efficiency while adding adaptive capacity.
  3. Experimental protocol: the manuscript supplies only the final LLM-optimized numbers on a single train/test split. No YOLO26 baseline with conventional tuning, no ablation removing the MoE module, and no multi-seed statistics are reported, so the contribution of each proposed component cannot be isolated.
minor comments (1)
  1. The title and abstract repeatedly use the term 'YOLO26' without clarifying whether this refers to an existing public release or a custom backbone; a brief statement of the base architecture version would improve clarity.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive and detailed comments. We have prepared point-by-point responses below and will revise the manuscript to improve reproducibility and clarity while preserving the core contributions.

read point-by-point responses

  1. Referee: Abstract: the central performance claim (0.9900 mAP@0.5, 0.9515 mAP@0.5:0.95, outperforming latest YOLO versions) is presented without any dataset description, train/test split, baseline implementations, ablation tables, or error bars. Because the attribution of these scores to the MoE module and LLM agent is the load-bearing assertion, the absence of controlled comparisons prevents verification of the claimed improvements.

    Authors: We agree that the abstract is concise and omits supporting details. In the revision we will add a brief description of the UAV insulator dataset, the 80/20 train/test split, and a statement that full baseline comparisons, ablations, and multi-seed statistics appear in the experimental section. The abstract itself cannot accommodate tables or error bars, but the main text already contains the supporting evidence for the performance attribution. revision: partial

  2. Referee: Method section (description of MoE integration): the sparse MoE module is stated to be placed in the high-resolution branch, yet no equations, routing function, expert count, or capacity factor are supplied. Without these details it is impossible to assess whether the modification is reproducible or whether it genuinely preserves one-stage efficiency while adding adaptive capacity.

    Authors: We acknowledge the omission. The revised method section will include the complete formulation: the top-2 routing function with softmax gating, eight experts, a capacity factor of 1.25, and the precise insertion equations into the YOLO26 high-resolution feature map. These additions will confirm both reproducibility and that the sparse activation keeps the overall inference cost comparable to the original one-stage detector. revision: yes

  3. Referee: Experimental protocol: the manuscript supplies only the final LLM-optimized numbers on a single train/test split. No YOLO26 baseline with conventional tuning, no ablation removing the MoE module, and no multi-seed statistics are reported, so the contribution of each proposed component cannot be isolated.

    Authors: We accept that additional controls are needed. The revised experimental section will report: (i) YOLO26 trained with standard grid-search tuning, (ii) an ablation variant with the MoE module removed, and (iii) mean and standard deviation over five independent random seeds. These results will isolate the contribution of the MoE insertion and the LLM-driven optimization. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical mAP results from standard training/evaluation

full rationale

The paper describes an architectural change (sparse MoE inserted into YOLO26 high-resolution branch) plus LLM-agent hyperparameter search, then reports mAP@0.5 = 0.9900 and mAP@0.5:0.95 = 0.9515 on UAV insulator imagery. These are measured outcomes of model training and test-set evaluation; no equations, first-principles derivations, or fitted parameters are presented that reduce the final scores to quantities defined by the same inputs. No self-citation chains, uniqueness theorems, or ansatzes are invoked to justify the performance numbers. The derivation chain is therefore self-contained and consists of ordinary empirical validation rather than any of the enumerated circular patterns.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 1 invented entities

The central claim rests on standard assumptions about YOLO-style detectors plus the untested premise that the added MoE block improves subtle-fault detection without efficiency loss; hyperparameters chosen by the LLM agent function as free parameters.

free parameters (1)
  • MoE routing and expert configuration
    Sparse MoE module parameters and routing weights are selected during the LLM-agent optimization process.
axioms (1)
  • domain assumption YOLO-family one-stage detectors remain efficient and accurate for small-object detection in UAV imagery when augmented with feature-refinement modules.
    Invoked implicitly when the authors state that the modification preserves one-stage efficiency while adding adaptive refinement.
invented entities (1)
  • YOLO26-MoE no independent evidence
    purpose: New detector variant that inserts a sparse MoE module into the high-resolution branch of YOLO26.
    Introduced in the paper as the core technical contribution.

pith-pipeline@v0.9.0 · 5777 in / 1332 out tokens · 66963 ms · 2026-05-20T05:38:49.610689+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

59 extracted references · 59 canonical work pages

  1. [1]

    Study on the mechanism and elec- trical characterization of the distribution porcelain insulator incipient fault in the non- effectively grounded system

    Cong, Z., Liu, Y., Yan, Y., Wang, K., Jiang, X.. Study on the mechanism and elec- trical characterization of the distribution porcelain insulator incipient fault in the non- effectively grounded system. IEEE Transactions on Power Delivery 2024;39(3):1840– 1851

    work page 2024

  2. [2]

    Enhanced insulator fault detection using optimized ensemble of deep learning models based on weighted boxes fusion

    Stefenon, S.F., Seman, L.O., Singh, G., Yow, K.C.. Enhanced insulator fault detection using optimized ensemble of deep learning models based on weighted boxes fusion. International Journal of Electrical Power & Energy Systems 2025;168:110682

    work page 2025

  3. [3]

    Nonintrusive ultrasonic sensing and deep learn- ing for outdoor ceramic insulator assessment

    Lutfi, A., El-Hag, A., Shaban, K.. Nonintrusive ultrasonic sensing and deep learn- ing for outdoor ceramic insulator assessment. IEEE Transactions on Dielectrics and Electrical Insulation 2024;31(6):2993–3000

    work page 2024

  4. [4]

    Evaluation of visible contamination on power grid insulators using convolutional neural networks

    Corso, M.P., Stefenon, S.F., Singh, G., Matsuo, M.V., Perez, F.L., Leithardt, V.R.Q.. Evaluation of visible contamination on power grid insulators using convolutional neural networks. Electrical Engineering 2023;105:3881–3894. 32

    work page 2023

  5. [5]

    If-yolo: An efficient and accurate detection algorithm for insulator faults in transmission lines

    Li, Y., Zhu, C., Zhang, Q., Zhang, J., Wang, G.. If-yolo: An efficient and accurate detection algorithm for insulator faults in transmission lines. IEEE Access 2024;12:167388–167403

    work page 2024

  6. [6]

    Uav-based automatic power line inspection using deep learning

    Zhang, R., Yang, B., Xiao, W., Liang, F., Liu, Y., Wang, Z.. Uav-based automatic power line inspection using deep learning. IEEE Transactions on Industrial Informatics 2022;18(3):1822–1832

    work page 2022

  7. [7]

    A review of power line inspection by unmanned aerial vehicle

    Liu, Y., Liu, J., Ke, Y.. A review of power line inspection by unmanned aerial vehicle. Journal of Physics: Conference Series 2021;1952(2):022014

    work page 2021

  8. [8]

    You only look once: Unified, real-time object detection

    Redmon, J., Divvala, S., Girshick, R., Farhadi, A.. You only look once: Unified, real-time object detection. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 2016, p. 779–788

    work page 2016

  9. [9]

    A lightweight insulator defect detection algorithm based on improved YOLOv5

    Chen, C., Zheng, Z., Xu, T., Guo, S., Feng, S., Yao, W., et al. A lightweight insulator defect detection algorithm based on improved YOLOv5. Sensors 2022;22(14):5226

    work page 2022

  10. [10]

    Automatic autonomous vision-based power line inspection: A review of current status and the potential role of deep learning

    Nguyen, V.N., Jenssen, R., Roverso, D.. Automatic autonomous vision-based power line inspection: A review of current status and the potential role of deep learning. International Journal of Electrical Power & Energy Systems 2018;99:107–120

    work page 2018

  11. [11]

    Power transmission line inspection robots: A review, trends and challenges for future research

    Alhassan, A.B., Zhang, X., Shen, H., Xu, H.. Power transmission line inspection robots: A review, trends and challenges for future research. International Journal of Electrical Power & Energy Systems 2020;118:105862

    work page 2020

  12. [12]

    To- ward automatic condition assessment of high-voltage transmission infrastructure using deep learning techniques

    Manninen, H., Ramlal, C.J., Singh, A., Rocke, S., Kilter, J., Landsberg, M.. To- ward automatic condition assessment of high-voltage transmission infrastructure using deep learning techniques. International Journal of Electrical Power & Energy Systems 2021;128:106726

    work page 2021

  13. [13]

    Uav-lidar aids automatic intelligent powerline inspection

    Guan, H., Sun, X., Su, Y., Hu, T., Wang, H., Wang, H., et al. Uav-lidar aids automatic intelligent powerline inspection. International Journal of Electrical Power & Energy Systems 2021;130:106987

    work page 2021

  14. [14]

    Outrageously large neural networks: The sparsely-gated mixture-of-experts layer

    Shazeer, N., Mirhoseini, A., Maziarz, K., Davis, A., Le, Q., Hinton, G., et al. Outrageously large neural networks: The sparsely-gated mixture-of-experts layer. In: International Conference on Learning Representations. 2017, p. 1–12

    work page 2017

  15. [15]

    Large language models as optimizers

    Yang, C., Wang, X., Lu, Y., Liu, H., Le, Q.V., Zhou, D., et al. Large language models as optimizers. arXiv:230903409 2024;3:1–42

    work page 2024

  16. [16]

    When large language model meets optimiza- tion

    Huang, S., Yang, K., Qi, S., Wang, R.. When large language model meets optimiza- tion. Swarm and Evolutionary Computation 2024;90:101663. 33

    work page 2024

  17. [17]

    Detection of power line insulator defects using aerial images analyzed with convolutional neural networks

    Tao, X., Zhang, D., Wang, Z., Liu, X., Zhang, H., Xu, D.. Detection of power line insulator defects using aerial images analyzed with convolutional neural networks. IEEE Transactions on Systems, Man, and Cybernetics: Systems 2020;50(4):1486–1498

    work page 2020

  18. [18]

    Insulator detection in aerial images for transmission line inspection using single shot multibox detector

    Wu, Q., An, J., Lin, B.. Insulator detection in aerial images for transmission line inspection using single shot multibox detector. IEEE Access 2019;7:9945–9956

    work page 2019

  19. [19]

    Insulator detection method in inspection image based on improved faster r-cnn

    Zhao, Z., Zhen, Z., Zhang, L., Qi, Y., Kong, Y., Zhang, K.. Insulator detection method in inspection image based on improved faster r-cnn. Energies 2019;12(7):1204

    work page 2019

  20. [20]

    Research on edge intelligent recognition method oriented to transmission line insulator fault detection

    Deng, F., Xie, Z., Mao, W., Li, B., Shan, Y., Wei, B., et al. Research on edge intelligent recognition method oriented to transmission line insulator fault detection. International Journal of Electrical Power & Energy Systems 2022;139:108054

    work page 2022

  21. [21]

    Key target and defect detection of high- voltage power transmission lines with deep learning

    Liu, Z., Wu, G., He, W., Fan, F., Ye, X.. Key target and defect detection of high- voltage power transmission lines with deep learning. International Journal of Electrical Power & Energy Systems 2022;142:108277

    work page 2022

  22. [22]

    A fault detection method for transmission line components based on synthetic dataset and improved yolov5

    Song, J., Qin, X., Lei, J., Zhang, J., Wang, Y., Zeng, Y.. A fault detection method for transmission line components based on synthetic dataset and improved yolov5. International Journal of Electrical Power & Energy Systems 2024;157:109852

    work page 2024

  23. [23]

    Faster R-CNN: Towards real-time object detection with region proposal networks

    Ren, S., He, K., Girshick, R., Sun, J.. Faster R-CNN: Towards real-time object detection with region proposal networks. In: Advances in Neural Information Processing Systems; vol. 28. 2015, p. 91–99

    work page 2015

  24. [24]

    Bidirection-fusion-yolov3: An improved method for insulator defect detection using uav image

    Yang, Z., Xu, Z., Wang, Y.. Bidirection-fusion-yolov3: An improved method for insulator defect detection using uav image. IEEE Transactions on Instrumentation and Measurement 2022;71:1–8

    work page 2022

  25. [25]

    Insulator detection and damage identification based on improved lightweight yolov4 network

    Han, G., He, M., Zhao, F., Xu, Z., Zhang, M., Qin, L.. Insulator detection and damage identification based on improved lightweight yolov4 network. Energy Reports 2021;7:187–197

    work page 2021

  26. [26]

    YOLOv7: Trainable bag-of-freebies sets new state-of-the-art for real-time object detectors

    Wang, C.Y., Bochkovskiy, A., Liao, H.Y.M.. YOLOv7: Trainable bag-of-freebies sets new state-of-the-art for real-time object detectors. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2023, p. 7464–7475

    work page 2023

  27. [27]

    Insulator defect detection with deep learning: A survey

    Liu, C., Wu, Y., Liu, J., Sun, Z., Xu, H.. Insulator defect detection with deep learning: A survey. IEEE Access 2021;9:2355–2368

    work page 2021

  28. [28]

    Insulator fault detection based on im- proved YOLOv4 algorithm

    Wang, B., Dong, M., Ren, M., Wu, Z., Zhu, Y.. Insulator fault detection based on im- proved YOLOv4 algorithm. Journal of Physics: Conference Series 2021;1871(1):012067

    work page 2021

  29. [29]

    Sparse mixture of experts enhanced trans- former architecture for short-term hydroelectric reservoir volume prediction

    Seman, L.O., Yow, K.C., Stefenon, S.F.. Sparse mixture of experts enhanced trans- former architecture for short-term hydroelectric reservoir volume prediction. Electric Power Systems Research 2026;255:112754. 34

    work page 2026

  30. [30]

    Scaling vision with sparse mixture of experts

    Riquelme, C., Puigcerver, J., Mustafa, B., Neumann, M., Jenatton, R., Susano Pinto, A., et al. Scaling vision with sparse mixture of experts. In: Advances in Neural Infor- mation Processing Systems; vol. 34. 2021, p. 8583–8595

    work page 2021

  31. [31]

    Multimodal contrastive learning with LIMoE: the language-image mixture of experts

    Mustafa, B., Riquelme, C., Puigcerver, J., Jenatton, R., Houlsby, N.. Multimodal contrastive learning with LIMoE: the language-image mixture of experts. In: Advances in Neural Information Processing Systems; vol. 35. 2022, p. 9564–9576

    work page 2022

  32. [32]

    Mixture of experts (MoE): A big data perspective

    Gan, W., Ning, Z., Qi, Z., Yu, P.S.. Mixture of experts (MoE): A big data perspective. Information Fusion 2025;127:103664

    work page 2025

  33. [33]

    Random search for hyper-parameter optimization

    Bergstra, J., Bengio, Y.. Random search for hyper-parameter optimization. Journal of Machine Learning Research 2012;13(1):281–305

    work page 2012

  34. [34]

    Practical Bayesian optimization of machine learning algorithms

    Snoek, J., Larochelle, H., Adams, R.P.. Practical Bayesian optimization of machine learning algorithms. In: Advances in Neural Information Processing Systems; vol. 25. 2012, p. 2951–2959

    work page 2012

  35. [35]

    Neural architecture search with reinforcement learning

    Zoph, B., Le, Q.V.. Neural architecture search with reinforcement learning. In: International Conference on Learning Representations; vol. 2. 2017, p. 1–16

    work page 2017

  36. [36]

    DARTS: Differentiable architecture search

    Liu, H., Simonyan, K., Yang, Y.. DARTS: Differentiable architecture search. In: International Conference on Learning Representations; vol. 2. 2019, p. 13

    work page 2019

  37. [37]

    Efficient neural architecture search via parameter sharing

    Pham, H., Guan, M., Zoph, B., Le, Q., Dean, J.. Efficient neural architecture search via parameter sharing. In: International Conference on Machine Learning. PMLR; 2018, p. 4095–4104

    work page 2018

  38. [38]

    Scaled-YOLOv4: Scaling cross stage partial network

    Wang, C.Y., Bochkovskiy, A., Liao, H.Y.M.. Scaled-YOLOv4: Scaling cross stage partial network. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2021, p. 13029–13038

    work page 2021

  39. [39]

    Language models are few-shot learners

    Brown, T., Mann, B., Ryder, N., Subbiah, M., Kaplan, J.D., Dhariwal, P., et al. Language models are few-shot learners. In: Advances in Neural Information Processing Systems; vol. 33. 2020, p. 1877–1901

    work page 2020

  40. [40]

    Mathematical discoveries from program search with large language models

    Romera-Paredes, B., Barekatain, M., Novikov, A., Balog, M., Kumar, M.P., Dupont, E., et al. Mathematical discoveries from program search with large language models. Nature 2024;625(7995):468–475

    work page 2024

  41. [41]

    EvoPrompting: Language models for code-level neural architecture search

    Chen, A., Dohan, D., So, D.. EvoPrompting: Language models for code-level neural architecture search. Advances in Neural Information Processing Systems 2024;36

    work page 2024

  42. [42]

    Large language model agent for hyper-parameter optimization

    Liu, S., Gao, C., Li, Y.. Large language model agent for hyper-parameter optimization

    work page

  43. [43]

    URL:https://arxiv.org/abs/2402.01881.arXiv:2402.01881. 35

    work page arXiv

  44. [44]

    Llmopt: Learning to define and solve general optimization problems from scratch

    Jiang, C., Shu, X., Qian, H., Lu, X., Zhou, J., Zhou, A., et al. Llmopt: Learning to define and solve general optimization problems from scratch. arXiv:241013213 2024;2:1– 27

    work page 2024

  45. [45]

    Survey of hallucination in natural language generation

    Ji, Z., Lee, N., Frieske, R., Yu, T., Su, D., Xu, Y., et al. Survey of hallucination in natural language generation. ACM Computing Surveys 2023;55(12):1–38

    work page 2023

  46. [46]

    Efenet: An efficient edge-feature enhanced network for power insulator detection

    Zhang, J., Chen, C., Lu, M., Li, L., Shen, Y., Du, J., et al. Efenet: An efficient edge-feature enhanced network for power insulator detection. International Journal of Electrical Power & Energy Systems 2025;171:111019

    work page 2025

  47. [47]

    Optimizing power system edge computing with a high-performance and light-weight yolo-based substation equipment defect detection network

    Wang, Q., Liu, R., Qin, S., Pu, J., Shi, R., Wang, Y.. Optimizing power system edge computing with a high-performance and light-weight yolo-based substation equipment defect detection network. International Journal of Electrical Power & Energy Systems 2025;172:111088

    work page 2025

  48. [48]

    Yolo26: Key archi- tectural enhancements and performance benchmarking for real-time object detection

    Sapkota, R., Cheppally, R.H., Sharda, A., Karkee, M.. Yolo26: Key archi- tectural enhancements and performance benchmarking for real-time object detection. arXiv:250925164 2025;2:1–14

    work page 2025

  49. [49]

    MFI-YOLO: Multi-fault insulator detection based on an improved YOLOv8

    He, M., Qin, L., Deng, X., Liu, K.. MFI-YOLO: Multi-fault insulator detection based on an improved YOLOv8. IEEE Transactions on Power Delivery 2024;39(1):168–179

    work page 2024

  50. [50]

    YOLOV9-CBM: An improved fire detection algorithm based on YOLOV9

    Geng, X., Han, X., Cao, X., Su, Y., Shu, D.. YOLOV9-CBM: An improved fire detection algorithm based on YOLOV9. IEEE Access 2025;13:19612–19623

    work page 2025

  51. [51]

    Ead-yolov10: Lightweight steel surface defect detection algorithm research based on yolov10 improvement

    Haoyan, H., Jinwu, T., Haibin, W., Xinyun, L.. Ead-yolov10: Lightweight steel surface defect detection algorithm research based on yolov10 improvement. IEEE Access 2025;13:55382–55397

    work page 2025

  52. [52]

    Rsm-yolov11: Lightweight steel surface defect segmentation algorithm research based on yolov11 im- provement

    Shan, Z., Haoyan, H., Zhu, C., Du, S., Jing, H., Haibin, W.. Rsm-yolov11: Lightweight steel surface defect segmentation algorithm research based on yolov11 im- provement. IEEE Access 2025;13:111681–111698

    work page 2025

  53. [53]

    Enhanced yolov12 through sliced contrastive supervision and full scene fine-tuning

    Garza, J.E., Islam, M.F.. Enhanced yolov12 through sliced contrastive supervision and full scene fine-tuning. IEEE Access 2025;13:138813–138819

    work page 2025

  54. [54]

    Site landscape design optimization based on YOLOv13 and multi-task bayesian federated learning (YOLOv13-HyperACE-BFL)

    Lv, J., Fu, J., Shen, D.. Site landscape design optimization based on YOLOv13 and multi-task bayesian federated learning (YOLOv13-HyperACE-BFL). IEEE Access 2025;13:185689–185708

    work page 2025

  55. [55]

    Orientedformer: An end-to-end transformer-based oriented object detector in remote sensing images

    Zhao, J., Ding, Z., Zhou, Y., Zhu, H., Du, W.L., Yao, R., et al. Orientedformer: An end-to-end transformer-based oriented object detector in remote sensing images. IEEE Transactions on Geoscience and Remote Sensing 2024;62:1–16. 36

    work page 2024

  56. [56]

    Input attention, squeeze and excitation, and spatial transformer of yolo for fault detection using uav

    Carvalho, J.P.M., Stefenon, S.F., Leithardt, V.R.Q., Seman, L.O., Yow, K.C., Santana, J.F.D.P.. Input attention, squeeze and excitation, and spatial transformer of yolo for fault detection using uav. Ain Shams Engineering Journal 2026;17(3):104067

    work page 2026

  57. [57]

    Switch transformers: Scaling to trillion param- eter models with simple and efficient sparsity

    Fedus, W., Zoph, B., Shazeer, N.. Switch transformers: Scaling to trillion param- eter models with simple and efficient sparsity. Journal of Machine Learning Research 2022;23(120):1–39

    work page 2022

  58. [58]

    Ultralytics YOLO evolution: An overview of YOLO26, YOLO11, YOLOv8 and YOLOv5 object detectors for computer vision and pattern recognition

    Sapkota, R., Karkee, M.. Ultralytics YOLO evolution: An overview of YOLO26, YOLO11, YOLOv8 and YOLOv5 object detectors for computer vision and pattern recognition. arXiv:251009653 2025;2:1–16

    work page 2025

  59. [59]

    Insulator defect detection.https://ieee-dataport.org/ competitions/insulator-defect-detection; 2021

    Lewis, D., Kulkarni, P.. Insulator defect detection.https://ieee-dataport.org/ competitions/insulator-defect-detection; 2021. Accessed on January 18, 2026. 37

    work page 2021